Analysis of Urban Changes in High-resolution Remote Sensing Images Based on the Improved ResNet Model

"The overall urban planning of Beijing (2016-2035)" proposed "reduced development," which is highly concerned about the existing stock and highly sensitive to development variables. Facing the demand for the rapid discovery of changes in information regarding urban land cover elements, we make full use of the existing image and vector data resources accumulated over many years to carry out research on the discovery of urban change based on deep learning. To address the problems of low accuracy and poor anti-noise ability of the existing methods for the detection of changes in remote sensing images, a method for detecting change based on an improved Residual Network (ResNet) is proposed. By introducing a channel attention module, this method can make the network focus on information from the specific area of change in an image, thereby more efficiently completing the extraction and reconstruction of the features of a specific change. The effectiveness and reliability of this method are verified using a sample set based on the Beijing No. 2 image. By this method to achieve automatic all-element change polygon extraction, the accuracy, recall, and F1 are all above 85%, which is better than other models, enabling the rapid discovery and accurate location of urban spatial changes and providing strong technical support for innovative urban spatial monitoring and modes of supervision.